llama : fix FA when KV cache is not used (i.e. embeddings) #12825
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
github-actions
bot
added
examples
python
ngxson
approved these changes
Apr 8, 2025
tastelikefeet
added a commit
to tastelikefeet/llama.cpp
that referenced
this pull request
Apr 10, 2025
* master: (123 commits) cuda : add f32 to bf16 copy op (ggml-org#12806) llava: improve clip_ctx destructor to not memleak load_image_size (ggml-org#12834) llama : fix FA when KV cache is not used (i.e. embeddings) (ggml-org#12825) server : fix thread.join() on exit (ggml-org#12831) llava: add more helper functions to check projector types in clip context (ggml-org#12824) arg : Including limits file on AIX (ggml-org#12822) server : webui : Improve Chat Input with Auto-Sizing Textarea (ggml-org#12785) Revert "sycl:remove redundant memcopy in function ggml_backend_sycl_buffer_set_tensor" (ggml-org#12812) gguf-py : support lazy tensor splitting (ggml-org#12809) llama : Support llama 4 text-only (ggml-org#12791) opencl: better identify Adreno GPU (ggml-org#12760) hellaswag: display estimated score confidence interval (ggml-org#12797) cuda : fix HIP and MUSA BF16 (#0) sync : ggml ggml : simplify Arm fp16 CPU logic (ggml/1177) CUDA: don't convert BF16 weights to FP32 (ggml/1174) cpu: move all the operators into a separate c++ file (except mul_mat) (ggml/1167) sycl: remove redundant memcopy in function ggml_backend_sycl_buffer_set_tensor (ggml-org#12734) ci : no curl on ggml-ci (ggml-org#12796) cmake : enable curl by default (ggml-org#12761) ... # Conflicts: # common/arg.cpp # common/common.cpp # common/common.h
colout
pushed a commit
to colout/llama.cpp
that referenced
this pull request
Apr 21, 2025
…12825) * ggml : FA supports F32 V * graph : cast KV to F16 when the KV cache is not used ggml-ci * server : add test that exercises embeddings with FA enabled ggml-ci
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
fix #12815
When computing the attention and the KV cache is not used (e.g. for embedding models with non-causal attention) the FA branch didn't cast the K and V tensors and they remained in F32 format. We now cast them to F16 to avoid adding full FA support for F32 types.
Also:
TODO:
We still allocate the KV cache for embeddings models, even though it is not used by the computation graph. This should be fixed by refactoring
llama-contextto allowkv_selfto not be constructed. This would save a lot of VRAM and will improve the embedding performance because we will no longer search for KV cache slots for each batch. Might try to fix this in #12799.